Quantum machine learning using atom-in-molecule-based fragments selected on the fly.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Materials Science

Background:

• Exploring chemical space for new molecules and materials is crucial but computationally expensive.
• Current quantum chemistry methods face limitations in comprehensive in silico screening due to high computational costs.

Purpose of the Study:

• To develop a computationally efficient and accurate method for exploring chemical space.
• To overcome the limitations of traditional in silico screening methods.
• To design novel molecules, materials, and experiments.

Main Methods:

• Combining atom-in-molecule-based fragments ('amons') with active learning.
• Developing transferable quantum machine learning (ML) models, termed AML models.
• Validating AML models for molecular quantum properties like energies, forces, and charges.

Main Results:

• Demonstrated efficiency, accuracy, scalability, and transferability of AML models.
• Successfully applied AML models to diverse systems including organic molecules, 2D materials, DNA base pairs, and proteins.
• Showcased the ability to reconstruct chemistries from local building blocks, extending Mendeleev's table concept.

Conclusions:

• The AML approach offers a significant advancement in computational chemistry.
• This method enables systematic exploration and design of chemical compounds and materials.
• The approach effectively accounts for chemical environments, facilitating the reconstruction of complex chemistries.